Make entropy context scalable to atom transform block size
This commit makes the entropy context operation scalable to the
atom transform block size. No coding statistics change is incurred.
Change-Id: Id10161116d297a2e65815e115993b79542e5a031
diff --git a/av1/common/entropy.h b/av1/common/entropy.h
index 43f849f..8d2a71f 100644
--- a/av1/common/entropy.h
+++ b/av1/common/entropy.h
@@ -218,13 +218,58 @@
const ENTROPY_CONTEXT *l) {
ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
- switch (tx_size) {
#if CONFIG_CB4X4
+ switch (tx_size) {
case TX_2X2:
above_ec = a[0] != 0;
left_ec = l[0] != 0;
break;
+ case TX_4X4:
+ above_ec = !!*(const uint16_t *)a;
+ left_ec = !!*(const uint16_t *)l;
+ break;
+ case TX_4X8:
+ above_ec = !!*(const uint16_t *)a;
+ left_ec = !!*(const uint32_t *)l;
+ break;
+ case TX_8X4:
+ above_ec = !!*(const uint32_t *)a;
+ left_ec = !!*(const uint16_t *)l;
+ break;
+ case TX_8X8:
+ above_ec = !!*(const uint32_t *)a;
+ left_ec = !!*(const uint32_t *)l;
+ break;
+ case TX_8X16:
+ above_ec = !!*(const uint32_t *)a;
+ left_ec = !!*(const uint64_t *)l;
+ break;
+ case TX_16X8:
+ above_ec = !!*(const uint64_t *)a;
+ left_ec = !!*(const uint32_t *)l;
+ break;
+ case TX_16X16:
+ above_ec = !!*(const uint64_t *)a;
+ left_ec = !!*(const uint64_t *)l;
+ break;
+ case TX_16X32:
+ above_ec = !!*(const uint64_t *)a;
+ left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
+ break;
+ case TX_32X16:
+ above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
+ left_ec = !!*(const uint64_t *)l;
+ break;
+ case TX_32X32:
+ above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
+ left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
+ break;
+ default: assert(0 && "Invalid transform size."); break;
+ }
+ return combine_entropy_contexts(above_ec, left_ec);
#endif
+
+ switch (tx_size) {
case TX_4X4:
above_ec = a[0] != 0;
left_ec = l[0] != 0;
diff --git a/av1/encoder/rd.c b/av1/encoder/rd.c
index 2797ebc..fb4960d 100644
--- a/av1/encoder/rd.c
+++ b/av1/encoder/rd.c
@@ -562,10 +562,83 @@
const ENTROPY_CONTEXT *const left = pd->left_context;
int i;
- switch (tx_size) {
+
#if CONFIG_CB4X4
+ switch (tx_size) {
case TX_2X2:
+ memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
+ memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
+ break;
+ case TX_4X4:
+ for (i = 0; i < num_4x4_w; i += 2)
+ t_above[i] = !!*(const uint16_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 2)
+ t_left[i] = !!*(const uint16_t *)&left[i];
+ break;
+ case TX_8X8:
+ for (i = 0; i < num_4x4_w; i += 4)
+ t_above[i] = !!*(const uint32_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 4)
+ t_left[i] = !!*(const uint32_t *)&left[i];
+ break;
+ case TX_16X16:
+ for (i = 0; i < num_4x4_w; i += 8)
+ t_above[i] = !!*(const uint64_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 8)
+ t_left[i] = !!*(const uint64_t *)&left[i];
+ break;
+ case TX_32X32:
+ for (i = 0; i < num_4x4_w; i += 16)
+ t_above[i] =
+ !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
+ for (i = 0; i < num_4x4_h; i += 16)
+ t_left[i] =
+ !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
+ break;
+ case TX_4X8:
+ for (i = 0; i < num_4x4_w; i += 2)
+ t_above[i] = !!*(const uint16_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 4)
+ t_left[i] = !!*(const uint32_t *)&left[i];
+ break;
+ case TX_8X4:
+ for (i = 0; i < num_4x4_w; i += 4)
+ t_above[i] = !!*(const uint32_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 2)
+ t_left[i] = !!*(const uint16_t *)&left[i];
+ break;
+ case TX_8X16:
+ for (i = 0; i < num_4x4_w; i += 4)
+ t_above[i] = !!*(const uint32_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 8)
+ t_left[i] = !!*(const uint64_t *)&left[i];
+ break;
+ case TX_16X8:
+ for (i = 0; i < num_4x4_w; i += 8)
+ t_above[i] = !!*(const uint64_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 4)
+ t_left[i] = !!*(const uint32_t *)&left[i];
+ break;
+ case TX_16X32:
+ for (i = 0; i < num_4x4_w; i += 8)
+ t_above[i] = !!*(const uint64_t *)&above[i];
+ for (i = 0; i < num_4x4_h; i += 16)
+ t_left[i] =
+ !!(*(const uint64_t *)&left[i] | *(const uint64_t *)&left[i + 8]);
+ break;
+ case TX_32X16:
+ for (i = 0; i < num_4x4_w; i += 16)
+ t_above[i] =
+ !!(*(const uint64_t *)&above[i] | *(const uint64_t *)&above[i + 8]);
+ for (i = 0; i < num_4x4_h; i += 8)
+ t_left[i] = !!*(const uint64_t *)&left[i];
+
+ default: assert(0 && "Invalid transform size."); break;
+ }
+ return;
#endif
+
+ switch (tx_size) {
case TX_4X4:
memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
